Full implementation of "End-to-end microphone permutation and number invariant multi-channel speech separation" (Interspeech 2020)

• Data pre-processing
• Training
• Inference
• Separate

First, you have to generate dataset from followed link.

Data generation script: https://github.com/yluo42/TAC/tree/master/data

You can use our code by changing data_script/tr.scp, cv.scp, tt.scp as your data directory.

# In scp file

D:/MC_Libri_fixed/tr # your path
20000 # the number of samples

Second,

python train.py

Third,

python evaluate.py

https://github.com/yluo42/TAC/

We achive SI-SNRi 11.36 dB in 6 microphone noisy reverberant setting.